Research map / Vision-language and multimodal / Robot action and navigation
Dive into the Scene: Breaking the Perceptual Bottleneck in Vision-Language Decision Making via Focus Plan Generation
Boyuan Xiao, Bohong Chen, Yumeng Li, Ji Feng, Yao-Xiang Ding, Kun Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the two-stage focus plan (SceneDiver) to the VLM | Qwen2.5-VL-7B · Robot manipulation (MuJoCo) | success rate %: 14.7 -> 28.7 | Improved |
| add SceneDiver focus | gemini-2.5-flash · Robot manipulation (MuJoCo) | success rate %: 38.7 -> 46.7 | Improved |
| add SceneDiver focus | Qwen2.5-VL-7B · Room navigation (Base) | success rate %: 32.7 -> 44.0 | Improved |
| add SoM visual prompting instead of SceneDiver | Qwen2.5-VL-7B · Room navigation (Base) | success rate %: 32.7 -> 30.0 | No clear effect |
| add extended thinking mode to the VLM | gemini-2.5-flash · Room navigation (Base) | success rate %: 68.0 -> 70.0 | No clear effect |
| add SceneDiver adapter to OpenVLA-OFT | OpenVLA-OFT · LIBERO-plus Robot Initial States (Spatial) | success rate %: 19.91 -> 29.49 | Improved |
| add SceneDiver adapter to OpenVLA-OFT | OpenVLA-OFT · LIBERO-plus Long, Light Conditions | success rate %: 84.47 -> 92.42 | Improved |
| add SceneDiver adapter to OpenVLA-OFT | OpenVLA-OFT · LIBERO-plus Object, Objects Layout | success rate %: 76.01 -> 76.82 | No clear effect |
| add coarse-to-fine focus planning instead of scene graph as text only (ablation) | SceneDiver · Room navigation (Base) | success rate %: 30.7 -> 44.0 | Improved |
| add step-by-step coarse-to-fine planning instead of direct focus (ablation) | SceneDiver · Room navigation (Base) | success rate %: 34.0 -> 44.0 | Improved |
| add fine-grained verification stage to coarse planning (ablation) | SceneDiver · Room navigation (Base) | success rate %: 37.3 -> 44.0 | Improved |
| drop 10% to 50% of scene graph nodes (ablation) | SceneDiver · Room navigation stress test | recovery rate: 96.78 -> 97.69 | No clear effect |
Closest papers by meaning
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 2026 · similarity 0.90)
- CoFFT: Chain of Foresight-Focus Thought for Visual Language Models (NeurIPS 2025 · similarity 0.90)
- Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.89)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- Evaluating Vision-Language Models as Evaluators in Path Planning (CVPR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
